Initial commit

code/keras_contrib/layers/advanced_activations/swish.py

+from keras import backend as K
+from keras.layers import Layer
+
+
+class Swish(Layer):
+    """ Swish (Ramachandranet al., 2017)
+
+    # Input shape
+        Arbitrary. Use the keyword argument `input_shape`
+        (tuple of integers, does not include the samples axis)
+        when using this layer as the first layer in a model.
+
+    # Output shape
+        Same shape as the input.
+
+    # Arguments
+        beta: float >= 0. Scaling factor
+            if set to 1 and trainable set to False (default),
+            Swish equals the SiLU activation (Elfwing et al., 2017)
+        trainable: whether to learn the scaling factor during training or not
+
+    # References
+        - [Searching for Activation Functions](https://arxiv.org/abs/1710.05941)
+        - [Sigmoid-weighted linear units for neural network function
+           approximation in reinforcement learning](https://arxiv.org/abs/1702.03118)
+    """
+
+    def __init__(self, beta=1.0, trainable=False, **kwargs):
+        super(Swish, self).__init__(**kwargs)
+        self.supports_masking = True
+        self.beta = beta
+        self.trainable = trainable
+
+    def build(self, input_shape):
+        self.scaling_factor = K.variable(self.beta,
+                                         dtype=K.floatx(),
+                                         name='scaling_factor')
+        if self.trainable:
+            self._trainable_weights.append(self.scaling_factor)
+        super(Swish, self).build(input_shape)
+
+    def call(self, inputs, mask=None):
+        return inputs * K.sigmoid(self.scaling_factor * inputs)
+
+    def get_config(self):
+        config = {'beta': self.get_weights()[0] if self.trainable else self.beta,
+                  'trainable': self.trainable}
+        base_config = super(Swish, self).get_config()
+        return dict(list(base_config.items()) + list(config.items()))
+
+    def compute_output_shape(self, input_shape):
+        return input_shape